物件導向式資料庫系統叢集及暫存區管理之模擬研究

Abstract

在物件導向式資料庫系統，物件是分散儲存在磁碟上，而叢集相關物件在 相同的磁碟區是改善系統存取效益的重要技巧。此外，緩衝區管理也是另 一個重要的效益改善技巧，藉儲存經常存取之資料於緩衝區以減少進出磁 碟次數。在這篇論文，我們建立一個模擬模型並且對叢集和緩衝區管理做 整合的模擬研究。我們並且提出一種新的叢集演算法與其它的叢集演算法 在多種查詢型別下做效益的比較。根據模擬的結果，所提的叢集演算法對 混合的查詢型別在固定的叢集路徑有最好的效能，而在不同的叢集演算法 下，LRU是合適的取代策略。

In object oriented database systems, objects scatter around disk pages. Clustering related objects is an important technique for improving system access performance. On the other hand, buffer management is another important performance issue, as it caches the frequently accessed data in the buffer pool to reduce disk I/O cost. In this thesis, we construct a simulation model and perform a simulation study of the integration effect of clustering and buffer management. In addition, we propose a new clustering algorithm and compare its performance with other clustering algorithms under various query types. According to the simulation results, the proposed clustering algorithm shows the best performance for mixed query types in a fixed clustering path, and LRU is an appropriate replacement policy under different clustering algorithms.